KEMBAR78
MLT 1 Unit Notes | PDF
0% found this document useful (0 votes)
291 views16 pages

MLT 1 Unit Notes

1) Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human involvement. 2) There are three primary approaches to machine learning: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning involves labeled data and is used for classification and regression. Unsupervised learning uses unlabeled data and looks for hidden patterns or grouping in the data. Reinforcement learning involves learning through trial-and-error interactions with a dynamic environment. 3) Machine learning is being applied across many fields to solve complex problems, make predictions, and develop intelligent systems using large datasets. It

Uploaded by

saksham
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
291 views16 pages

MLT 1 Unit Notes

1) Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human involvement. 2) There are three primary approaches to machine learning: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning involves labeled data and is used for classification and regression. Unsupervised learning uses unlabeled data and looks for hidden patterns or grouping in the data. Reinforcement learning involves learning through trial-and-error interactions with a dynamic environment. 3) Machine learning is being applied across many fields to solve complex problems, make predictions, and develop intelligent systems using large datasets. It

Uploaded by

saksham
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

LT Onit-

ntaoducti oh

achihe Lea nina is asubtieid Ar+if icta) dntelligent


the
AT)Te Joa o ML gene»a IS to urdes s t a d

modeiS t h o t
Struct se odoala and i t tht dorta i o

Ccn be understood aned utilizCd b People


ML alqoait hms dilos tan on clacH Ca inprts E ose
Computeas to

Statistical Cncaly siS in odde to ol ev al ue that t a l wdhih


Specii Tane, Becaue o this m achihe leaahina tacilHaey
in builelina mocels o m S amle l a t a in oader
Computeos
to automate bayel dat
de cision-mak" Pece ses on

ihputs
e Facial zeloaniion echhol0) allows Social meola

Platovns to helP used s + shar photos tjndd


OPtial Chavcte necogihianOCR) Teckolo coV&t
e s o text int mvable ty oer Ke commenolation enqih s

POweaed b mL SUggest hat mvies 0.3+V shows to

a t c h nert based on Uses P>ee>0 Ces S e t daiuin

CoaS aela o n ML o nauide maY Sooh be auulablo


ConsuMess

hes or_ML

ML

SupeR UsE
LE ARMINO UNSDPERU*SE O KEINFORCE HEHT
TASK0QVEN) LE ARHIN LE RRNN
DATA-ORwEN)
(Leon o m a r s )
Nole La beled Vata t unlabeled Vat
obele d Vala ) Labeled olarta hos oth he nput outbut Pasameie8S
Po Conbie +e IC
d eadabl kattexh, but Aeuies lo+ot hun ah
lo bo to lobel the data
to begin coih,
Un labeled dada nlabeled dacta ond has o ne 0d enon e ot he
ParaneteBS n a hachine ->e adaHe tearm TRS
hegates the he
ed humoh labo bal aequ>e e Conplex

SolutionS
Subeau ise d Leaanir)- Supeised leavninais one +he most
ML Ln +hS t e , h e ML
aljo s >alh eck
on labeled data
e t r e m e l poueaful e h used in nigt CicUnstahto
n Subezuised eadhin),the HL al90 is ven a Small
acxininq dortase-t Cooak itk, hisa-aih dctcaset is a
Smallea baat o+he g e dat cse+
ahd seUes
serues +o ee the
ane
ayoo busic idea OT
paobem,Seution, aho olata pointS T
be dealtt wth. T *ahin doOSet iS also v Qay Smilad
tot 4inal dataset n its chaae> \St S anol baowidas
e alo CTh holakele ol Pa-ahete>s >equireelto
th
Te a then inds e l«hionshm1bs betwn
t e paamele
ive egertta AL he end o he taai.
+he aljo has om hos+he

Melationships ble ke cndd the Ou


daployeo to> wne

th+e Aloart a e t , udrtr


machine Leaanh
woak th Uhbbelec
holds +ie adlvontae t bein able to
m@S that hum&n a b od is not e qi e c l t O
dada
hackin-AaCable allouin much laag
mako He data
6e
oePd on b +e Paoaum. Unspeta{eA
utas ets

ecn oes mot huve \uhes


o or eulthng in

Stzutuss RelhbnshKis b l data fat


Ceahon or hidcdlon
,

mahhe,
ObL per Ce b alydaithm in a n abstaact
e
ith ho ind Cqel fsom homan " &har koy
1S
hke
e caedon r nicodan t
Hae tae

USueuSea leanin aso versatile.


eanh diectt
I P a h h divectiy
Pein toa tenekt
3) aintozcement eaann-
CKe inSpiTethon mhow homa bei hji tecunhfaO
Cn algoaith at
dot a in tHeir \eS, At eaAuae)

mpooveS on itse ond leaans tdo he stuctions USi


aeen(oudagaL
*ial-dhd-eaor methoo.Favoaable ouHpats
honavozable outpuots a n e dustouaHEA
Or entorced, oJnd

O PonisLecl".

wel Posed a o n

machine
learh In ]
e s bein oua stud
Consi desi a es earh +a t*o
learh
Saidts
is
betion A Combuter Pooga
+asks |
t h espeei to Some claJS
C»Realence E
s s peao»-mahte
peaomancQ

Ait
erthertarane mea suwe
inp>oues H h
T s heasuaed by

epeiehte..
Paoyaom thad leaahs to Pla
tod e* A comþutez
CLS mea
qj)u N
cus m v a

ts fefo>mahnee
Cecke»S miaqht mprove
clasS G Ttt
aass
ks
attLe
bt ssabi to i n
epesêhe
expehe
Play
checkers anes, thao 3k
involi
itset.
Obtoineo
ba puaJina e s ajant
a well-cdeHhan \ead h
nnal t o haue
n hone e g e aduno
ncQeatune
ldonta
aoblem,ve must Pertor m othte
Pertorhhe
aks, e
meayude o
e class o
o bempaovek
e2 nd the Souce eniehte
to
A Chackes leanhiny paoble
Chetce>S
oTask eec Playin
efomoee Meusuae)rercent e wor QJuFY opeis
Tain eyretenceE pluuing aaice ne
itself
ouhst
Coe Con nsPoblemy like
ghect h
houncduaiti necoqntion leoahiny aobiem
Tas T Ae cJhigi anl Clasyl4ytn hed atlen
do vds thih ima
er onahte me e : ot ody
pe cert
Codect Cla sstted

Traniko
roni evpenience E a datrbcy e st Aandortt
wood Cotk 8tuer Classificottoh
Holoton shle Ahon a AML & DL

OLisa Ssbpazto HL

ML Subbat o
is Su AL AT
ML
Al
0L
Call ML alGo ed o AL

but nst Ce vevSa


Some dmpotect tezms

1 esaesjon
2 CLusAedin
3)Naive Bap
C
Cnce bt leanin
TSSu in M C Le onih

7Uhehohs
taaget
1) hod ago e*ist0 leaahi ere aal
Specitic +ouininCeamples seti 3 w
An uhat

P aatua atjo Conueage to the desiaed unetioh, Je h


okch alo Pcto m bes o7
Suthcient tadinira dlata9

Cohch pes r Paosle S 2 Jnee esemcetonS

much +aoining data suficiet h Ge a


Ho
How
+eleashe

houw Can prlos knolede k e l d 6


Chd
3 ) s t hen Can piey
ereradizi o n Coambles
+He parocexs o

helpjul Cen k a n it Sohl aypdooim atey


Khosled be

Codret
choosingautetuU hert
Cohat is e best e tracteja tor
) choice dr
now coes e
toni epeni ere, quh
leaanth Probem
a he
ale heCo mpleii
imteLtto
the e a a h
S ohat is He best ats e dute
hoe tuhctio h appro»imeeti oh oblem
hact
kat
0d tto
o leadh
One
Sstem aHem
lead
Stecitic
tuntionshuld the
tset be outomale
Caunhis pr ocess
ow C an the lec automatiraQu al ten tS

JepneetohOn to mProvE cLbiit o


>e pseIETt o h

easr t e tage Tuncthon


C humbe oT
Mg4e- Vesigsin8 a ml apPz oach
the tybe d t>ainihA
design Choice s, including Choosin eþaeserldion
to koe leadhe a
Cepeaiorce, the a et{untion

t a &untti0n,
nd au algoritm foy leayhit
eumpl0
heahinihvdve Seenck'. - Searching hog h a SpaE OT
o inol He kybothO)e thct be
Possible hypotheseg
an Ptos
its
ts hea awulae +>ain in eeanele ad ot
ConstaCunt S o8 knoole clge
n i e t leadnin
unthion3o taaini
Tnte
hteb CL boolecun-ValuQ

oletined ovor
Ceanples ot i olp.e A a Sunetion and-au
bird
animuls ohose val s zue o
Sor Othe nmuls),
A Concet leodhim

4e
value t Enuy Shut
Pre clitH
to otha
TRe oi S
the va
TOd n Gøbrtrun da, e c on

Sha00 D e
Sha0Q w
atHibutes a r e a d a t o n

Cnoot
hypothesic
Te y Pot hasiS
hpo-t il|
heiswil
n ths Ccae
Cee
Prov t o 4he. l e a n e a

e etthan hal an value scLCCeptable-


&
A cl'coct
+his atribute
+ha
Tor4ha
eg, wan)
w a n o r
e quurel value Cey,
value
Pecify a Sinyle
a t a i b u c e , Od
+had hó valu is acceptabl e
)Indit ade 6aP
alRE
Cill4he Consda
TSne instuhte DC
Scctir ie»
as a
a posiHve
bosiHve

h eh h ctaw tie as

of h p o t e is
eamplehd-1D
hoths nat Aldo enogs -his avcaite SPot d
Cith ighhumidlity ( Anclebe o
On Cold dos
th ohOn cutlAbetes) is aoPre)eria6
t h Vauas o o

ee poesio
,Cold, High.,,)
e n y SPoi
oind Cocclen oeCoy
Sko AT MU
No
C
Cho
Coo Ch

hu ev@ny d s a + ve
matgeresal ky t kais
2CA e C,1,1,,,2)
bossible hypeihajs-
unche hOst sþecific

KypDthesi
Mavimally Sheciti'c
INO- S+inddina
a

he tauinina ddta
bothesis hat
S earchor h Sbeciic
Sbeutic
hy þo thesis in M
the mostt
Italize
hto
nstance oC
rainin
each po sithve h
Cach
cttbibute Const>ant a in
.tor

the Const aant C IS SaHisiied bjo


Then do no4hin
byhe he mdeeheda
E I S eeplace a

Sqisicd by c
Constaunt hat s

3Output hPothesis h
in S A1O S ien He
To uStout e +his alA, assume he lean ex
table -
SeqenCe
OTtduthik erqmples toom bele
Eamble SK Ai lemp Mwnielt cdater tOpe cat Efoj Shoat
Sumy Noamal
So Same
e
2 Sonn arm Migh Staond Cam es
ain Cold Styn aym Chanye NO
Sunh warm
Staron) Cool Chan
ositive neative tdainin e»amples To the tarqe- Contept
engo sPor
The +iat Steb INO-S is t o intializ e h t o the most
SPe citC hyþothes in M1

2) oCning e»ampl e dom table hy Fve espom)e, so cch


'S eplaced o the heyt moye
Constunt that
enerul
Gits
t S the e»ctmble .
Si/ (Sunmy, ah, Noynal,St>oy,oat, Sume)
This abo
abo veve pothasis is Si ve Spe ci i C , 14 Say3
Ahad

a insBahceS ade hegative e c e pt o +he SinJie positve


scinin epamble a e dose>wed.
3) S2 Sun wam , , ston o a r o Same
m
Os tmid +ainin eample is
S nejative -
the ago
makes ho to h. At Simply ho&Es Cve hejahve
hcnge
enompies
S%/he-(Sunm joadm,,st>onj
A) S nd -S, t+k
SrtnolS, ke (+e )
foth (te
trace
) To
1o Comblete. o ehealitation
h
uther
ambla Ieads to a
,2)
h Sunna, Dam,, St>ong,
Compe ie Sol4
ho d , d,d, d, db77
Sunny, ,Normal, staony ,wam , Samee/
Wam

hi: Sunn, ATm "l S*on wadH, Same /


h-Sunndwadm/ Stoon ' ) a r m , Sam e
Sunn), w am, l, s taon ,?7
Tssue in1 e cSion T&ee Lecanin

tollo oin CAe ho maln SSues-

Aoiding OueriHin the data


Continuouy -ualued AH»ibuteg
)Ancoposatin
Select in Ataibut@
3) Altea native Heasuaes us
AHaibute Walues
9)-Hondlir Taalnin Eamples t h Missin
s) Handl lin Attaributes Ah Hesin Cots
include
+eeS
olecisioh
actica ISSues in Vee leaznina
taee
decisioh
l e t e minih hou deeb to Joa,tho

an appoo priate
handlin Continuo us Gttaibutes,
SChoosi n
data t h
Ottaibute &electon measuae, handling toaining
att-ibute corth
miSSin attatbute ualues, hancllin
COmpuiatonal eticiehc
Per n Costs and impaoo
hoise in
)Aoi olin Oveatin Ae 7 ata.- h e n t+hese S
examPes
the data Ov hen the numbe ni
OT
O e p a e s e n t a t ive Sampie
S too Smull topaocute

t Cun leud to diHiculties


\he tTue t a r e t tunction, Cuh
algoihm
OTthese cases, this Simple
ethe & amþles
oveiit the training
toeeeSs t h a t
paodUCe
ovetits training
the
CL hypotheseS
that
c an
Can Say
e
othe hypothesis that the
Some

Cam ples peroms


pertoms be
b e-ttlle
e

eanples iess Well actua


CLctuql
taning C instan ee
oT
Ces
instan Ces
clistaibutio
tne
me entire
Ove
InstanLeS bejond
h e
the dan\n
tdo se).
ncludin
hEH
hpo4heSis Seue M the dain data
hoveTt

Smalea CadO h a b h Oves +he t > a i n g e a m T


hhaS
e o d Hah hove the entia e distoigtio
h a s Smalle
n s tanceS.
Cpprouc he s tO avoid oveafiHin
deacheS
) astop aoLdin he taee eaAT{ed, bet»e
o i n t heaet peec Ciassities anl
clata Aranin valicdatoh Set)
that callos he taee to oeit +he dal
)
) aapprooaches
a r o r estmeetion)
CLho then
n o s t -prune the tree. chi2

Cotihuous Valuec AHaibulePS


aInCospotatin dicaeie
take dh a
at&ibutes that
o
V3 1S ne staicted
Se+ o values
Cotinuous - v a l u e c
he
AS Spose e wish o in lude
the A7ani na exxmple das
taibute Tem in ceCibin

nthe learhin aK
Tennis
Te m 2 4he g e t ct>ibut e Play
48 6 2 80 9O
Tem o
Play Tenmi NO No es es es Nob
h
odues
Ae eut e,
a t h deshold C

&8o19o)2
85
CCL hehb be Computao t o each he
The
then he best Ccn be Seleco
Cardid ate attbues
Yleasua es i o Selecting Attabutes
3) Atea nati v e
bias n
he IG measuO thT aucoS
heoe S C natuda

arth Ualues Ge those with t e alues,


Cttaibtes a
As an @e CeamplG Consde>tlo atl >ibute Vat
humbe OT Possible values e
Conich has a vea
aae alore peatetA Ped icls -tke i e t
Maach 4, 13 ). | yote
dool node OThe tee 2
TOr- he

aCue io od) t aee OT deolh ohe ohich eo Xer\J C kasrie S+e


data, Vate a v e ve Poor edielo
e Adauninq
o nS e h nStance2S
tooget unttion Over
OT he
meaSuIeS
C cah be
me other
So O auoid

sec insHeadOr IG, e- inKaioQuin lan 1986)


The C aaio m e a SUGe Penalzes at buies Such aS

Vate inCoa poraihg team Calld Sbli+


hat S sensitve o hou boo ad ond u n otm
Cttàtbute SPts \he dctla

pi&Tcomafon S,A) = - 5 s9 1S
S
4) HodingTaainin comeles h Missina AtHibuheg
value S
2AIn Cestan Ccuses+ke auailable dat a mad be mSs
be m

Some cttaibudes

c+ul out Come Some Ctlaibuue,

values ane missina. We Ca docal th t


b
mo s t wed
CommonVale s ha
Otabwi place
SSi Uale
Monddin Aabute colh Heaih Costs
Sorn e anin a SKS A ims a n ( e
>ibuies
'rawe asociaed CosAS.

orC Cla sri medical diseases aHibutes s be

empe rotua e Sios esutt ise ,8loocl les esult


h e e attai buey ConsiSsis t CostS balh
monetor
monetoT CoSt Coyt to puhent Comt o-+.
we ( n
modit D3 Such as takes attaibutes ba
0
me a wur t h e Cost
nencdinjthat ecttaibule
ae mt diuide the iain
bythe Cost aT +he a tiaibcteF
thut ues(OSf atibutes ould be Poeteasel .

St0 o ML

HLS buinesS
aspect or
ah moden
mpolant modelS Ao

9nese0adCh. es alg0ai thms neuTal

Pojae'sively impaa i h
assis Computed Systems in

thair ertomante
model OT
ML iS, based o
in 1949 J
ed
htesat+on, The mode was C

Vonal od ebb titlecA The Odgahizodion o


in a book
heudon
on

Behauio. T t ocpoesents Mebb's the


neurons.
bla
C i t ement CommuiniCation a

Samuel di I8M developed


Adthuad
Samuel
Computer poam checkeis n19Sos.
Paina
initiated cohat is Called alpha- beta paunin
a num be2
- mehaS mS
Samuel deiqnecd
hatt
allowin his PTOTam tto
o become better dn
n e lsokack
Samuel Called dote leaznin ii
S s am

emembesed C pos tionS ithad aldedy Seen 2


Cormbineck this oth h valuey TT h e e aadtohetioh

Aathu ist Cane op thophaaje Machie


damuel ih

952,
ean
I n he cte igaos E easl Eos
Apti icialnelligence oLe secach hao pcused on ujin
Khtlede -basecl appronches 91ather t h c h algos
oyi Cal
This Caued C eRrution blw AT 2 ML.On ehh
ML hac beeh used. as aatinin paoramioa AT.
ecent machine Le adnin CwoS deined b stanord
Onive7Si1 he Science etin ConPuters t o

acttoithout bein eplici Pao ammeo


is
Seven Common ways +e osld o busin ess
CUrrent OSin Machine Leadhin
O

Ana n Sales V a t a

2) Real T ime Mobile Yeasonaliza lioh


3 a a u d Ve tection

4) Produc+Kecomme ndationS

LeaaLeaanin Manegement SystemnS


6) Vnamic Aci
Yici
4) Hatyal Langua Proc e s i n

Vato Science vS T l a c h i n e Leaning


Leanig
can sayhut
sine OdS we
Vat a Science In dat a
data
Hhat
dhat

boce>sin
a r a l y si s
Ahe
duta Science insihts.too instance,

Ou enerate o v a d i OUS
on
esoa1eS.
Cot
eo products o
in Anrouh
hroujh C

dala or
dctaa oT pr oduct
ProduCt

Shop S t e s , Stoenesan
System
necomme ndahoh
baoC e s Cs o
Vaxt & Science nvolVeS the

) Vata Ctrachon

Vata Cie an s
3) Analysis
isuali zation
actonable nShts e r e aAl6h
s) SCience, dA ddaws
Ma chine Leanni ML post oT clala,
Onc LOthmS Ao O a K on the
aspectS Toomm Sta +iS+1CS
odlata qenerated ertraiecl 1aom multip1e 91esOUT ceS
the abilit ve to SYstemn le udh
Coithout human
PaOceSs daa Sets auto nomo usl
&
This S achievec hrough Combl ex ago
intesvention,
ClusteinN/ nalue
+echnlques ke Superuisec
Ae essioh ,

Buges moy
Tke
Te
C eund! date Elminatioh A\othm
h m
Cancdi d a t e -
Etimination A
The keg dea he
Se ot
elescauptio nd
+he
Comaut is o Outptt
deTa
S1Stmtr wTth t h e + r o J n n
au hypothesesS on
hr i h n

The veosion Space


denoted VSn,D
V Sn,0,
Weasion Space- eoamles O,
D,
spate M e+aining examples
ees
o s ect to
+o hy pothesis
o hhypothe
y ses A>on Con Si sten
Suhset
S
eamples n O
O.
Coith 4he taainin
he Cnsisten+(h,0J
VS, to
hypothe sis Space
eiespect
Cth
The qenesal boundaT , C
IT maimal °heaa
he Set
anin data 0, s
hy pothe sis
hy potheyis
me mbedS H Con sis tent to
despec
t h d e s p e c t t o

The Sbectic bouhdady SS, odh


Set a
mhimall
Cohi stes
he
decto 0, is
Soace M and+oain ing members ar
m a e i m a l l S b e c i i c )

8 eo c e C e , Sba c e
veS i
on

coth . ht the
we Can sho plos
paticola s G
In hypothYs
otain ol in

di
Prectye
Peciel
thSel
hal ie b
plus those
S
hos
o s ee Cont a ined ,
sPace
SPae
hy leohe
\he Sis
Ode eo hy eo
ahc h e eaatial
eadtialiy
Cand'date -

Etimina t ion algo s


in yeosion Spaey,NotIce
4he dualit Positive neatiue esam Ples inn luehce
S G

Intialvze C*o the Set maimall ehes al hypolhee

set o maimall Speitic hypelhese


to he
Inttalize

o euch JUun in eample d, do


IE d is C o sitiv e eeampC
w h ol
ih Consisten t
Kenov e zon G a n hy Potheses
hat S not Con si s ten+
in S
o each hypotheSISS
with
Remove saom S h st
qeneTalizations
> Add to Sall minimal

Sun n a
his Th o, hd some
Con sis tent
membe% C is moa e 2eh edal thaun
Kemove om y hypalheyis ht is

e
Jenesol than onoth e hPothe sis in
S a ne gahvee oample
emove om S and ho-thesis ihConsis Hent oih d

to eadh hypothesis in a thec s hot


Consistent with d
eme ve om n
Add to n call minimal SPecializccti' ch h
OT S Sucn+hot
h s Consista h , ancd
SCme membe &Sis
speittc tncn h.
oye

emeve oom Á an pothenis


hat S ess enea haun anethen hyrotkenis h
hyrotheis
o6,,,4,b47 Ce929921}
SSonywa m, oamal,Slaons, arl Sune
SSowAnashun, dasm, Sume? G:22,2,2227
S3Suy, tOamstaong demame} G3:<Sunna
2222

S sunn wam, , , strong,,


i2,22,I Save}
n {kSumm 2./22
1,S
, wa,2,2,77/
Sunny, wasm,..2,2,2>1, Sunn, 22,9taog,?,2
, waa, 2, Staora ?,13
Teire tox the Eyo S o r Conceb
deasion SPce lear nino

Pooblem +Bainn eamp es oesCibed eanlieA.

You might also like